Project 4 Computational Modeling of Motor Sequence Learning The overall goal of this project is to develop a biologically detailed computational model of learning in the discrete sequence production (DSP) task, which is the task that will be used in all projects of this PPG. The model, which will include multiple regions in premotor and motor cortices, as well as the basal ganglia, will integrate Ashby's work on category learning with Houk's distributed processing module model of the motor system. The final neural architecture used in the model will be based on empirical results from the other PPG projects. Even so, the basic architecture will include extensive cortical-cortical projections, and each cortical region will be connected to the striatum via closed loop pathways. Synaptic plasticity in cortex will be mediated by (2-factor) Hebbian learning, whereas plasticity at cortical-striatal synapses will be mediated by (3-factor) reinforcement learning (RL). Because of this difference, a fundamental hypothesis is that sequence learning in cortex requires initial assistance from the basal ganglia. The key idea is that the basal ganglia input to cortex serves as a critical scaffold for directing cortical plasticity. This approach has been highly successful in explaining category learning and will be extended to sequence learning in the current proposal. Aim 1 is to construct the model and test it against several qualitative benchmarks. These include verifying that the model can learn to make predictive responses (i.e., respond before the next visual cue is presented), and that it can eventually respond without assistance from the basal ganglia. Aim 2 will test the model against some classic published sequence-learning data. The final goal is to test the model that results from completing Aims 1 and 2 against data collected in the other PPG projects. In particular, the goal is for the same basic model to account simultaneously for single-unit recording data collected by Strick and Turner in Projects 1 and 3, for data from their muscimol inactivation experiments, for Strick's flavoprotein imaging data, for fMRI and TMS data collected by Grafton in Project 2, and also for behavioral data collected in all of these projects. Furthermore, a critical goal of the modeling will be to account for changes in all of these data types with the extensive training that is planned in each project. RELEVANCE (See instructions): The proposed work is central to the problem of understanding the mechansims where practice leads to to reorganization of the human motor system in the face of aging, neurodeneration, stroke or brain injury. Understanding these mechansims has an impact on the design of therapies directed at preserving function, developing compensator movements and ultimately, developing novel motor capacity.